Open TurtleZhong opened 1 month ago
Hi. The obj file is generated by 3Dscanner. In our paper, we use 3DGS to reconstruct 3D models.
Hi. The obj file is generated by 3Dscanner. In our paper, we use 3DGS to reconstruct 3D models.
thx. I found the pipeline in the Paper I noticed that the pipeline does not involve the 3D GS part. So, I need to take the provided COLMAP data format and perform 3D GS model reconstruction myself to obtain the final PLY file, correct?
Thank you for your assistance.
Correct.
Thats why I have not found any code about 3D GS? This part of code has not published yet.
@SophieZhou Hi. You can find 3DGS in their official repository. https://github.com/graphdeco-inria/gaussian-splatting
@SophieZhou Hi. You can find 3DGS in their official repository. https://github.com/graphdeco-inria/gaussian-splatting
Using the code from this link, I have trained my own model on your sample data using default parameters. But I got the rendering results a little poor, I do not know, what is the difference. Have you done any changes over gaussian splatting? The datasets are generated using your code and the models in your code step by step.
I downloaded the sample data and loaded the corresponding OBJ file using Blender. However, it seems inconsistent with what is shown in the paper. Below are the original image and the visualization of the OBJ file. It appears that the upper part of the vehicle is not well reconstructed, But the paper shows good results. Could you explain why this might be? Or is there any additional step I need to take to achieve a better 3D model?
The image like this: